Wyniki wyszukiwania - BazTech

1

Ocena modelu dyspersji zanieczyszczeń powietrza OSPM: studium przypadku, Polska, Kraków

Rzeszutek M., Bogacki M.

Rocznik Ochrona Środowiska

|

2016

|

Tom 18, cz. 2

351--362

EN

The study represents the evaluation of the OSPM model on the basis of the simulation results carried out for the street canyon located in the Zygmunt Krasinski alley in Krakow. Road emission was estimated based on the data from 2012 in regard to the traffic volume and the types of vehicles. Meteorological data were acquired from the station located in the vicinity of the examined canyon. Ambient air quality was determined based on the monitoring data from the urban background station in Krakow. The calculation results were compared with the 1-hour concentration measurements from the traffic station located on the green belt between the two lanes of the analysed canyon. The analysis was restricted to the nitrogen dioxide (NO2) and the PM10 and PM2.5 particulate matter concentrations. Model evaluation was carried out according to the methodology of the air quality models assessment. Values of OSPM model statistical evaluation parameters fall within the range of “good models”, which indicates a very good quality of this model calculation results. A very strong correlation between the results of calculations and the observations was found, particularly for the particulate matter PM10 and PM2.5. Correlation coefficient values for these pollutants are 0.90 and 0.91, respectively. A detailed analysis revealed that the overall quality of the model is better with respect to nitrogen dioxide than for the analysed dust pollutants. However, this model is burdened with a tendency to underestimate 1-hour concentrations of particulate matter, which can be associated with: negligence of the lifting effect of the dust deposited on the road surface, the effect of removing the particulate filters from vehicles and the adequate ambient air quality for the analysed monitoring site. Very good agreement of the simulation results and observations indicates that the OSPM modelling system can be successfully applied to support the air quality management system in Krakow.

2

Application of satellite observations for the evaluation of the land surface temperature from GEM model forecast

Gawuc L., Struzewska J., Kaminski J. W., Durka P.

Challenges of Modern Technology

|

2014

|

Vol. 5, no. 2

25--32

EN

Application of satellite observations for the evaluation of the land surface temperature from GEM model forecastAbstract: The Global Environmental Multiscale model (GEM) was evaluated against satellite observations and measurements from synoptic stations. The computational grid was set up in the global variable mode with the resolution of ~25 km over Central Europe. Model evaluation was performed over Central Europe within a window of 43-56°N latitude and 10-25°E longitude. Surface temperature forecasts were compared with the Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature product. Air temperature measured at the height of 2 metres was obtained from about 480 synoptic stations from 13 Central Europe countries. Air temperature measurements collected at 9 UTC and 12 UTC during five days (31 January, 2 February, 3 March, 27 April and 18 June 2012) was compared with the GEM model results. Evaluation showed good agreement between modelled and observed data. In case of air temperature, the averaged value of the Mean Bias Error (MBE) was -0.42, the averaged Root Mean Square Error (RMSE) and the Mean Absolute Gross Errors (MAGE) were 3.21 and 2.32, respectively. Land surface temperature comparisons gave results of -2.01; 3.91 and 3.24 of the (MBE), (RMSE), and (MAGE), respectively. Also, correlation of derived modelling errors between surface temperature and air temperature are discussed. In each case the correlation coefficient was positive. The highest value (0.70) was obtained for periods when surface – atmosphere radiative exchange processes were dominant.

3

On the identification of composite beam dynamics based upon experimental data

Dobrzański L. A., Honysz R., Fassois S.

Journal of Achievements in Materials and Manufacturing Engineering

|

2006

|

Vol. 16, nr 1-2

114--123

EN

Purpose: Article describes kinds and use procedures of mathematical parametric models describing dynamics of the systems based on excitation and vibration response signals. Design/methodology/approach: As a sample of identification of mathematical parametric models and estimation their parameters was a composite beam investigated under a white noise excitation force activity. Findings: Model based identification leads to finitely parameterised models described by differential equations. Research limitations/implications: Such models provide important features, in comparison with non-parametric systems: direct relationship with differential equation or physically significant modal representations used in engineering analysis, improved accuracy and frequency resolution, compactness/parsimony of representation. Practical implications: Ability to provide complete system characterisation by relatively few parameters, suitability for analysis, prediction, fault detection and control. Originality/value: Article is valuable for persons, that are interesting for identification of mathematical parametric models and vibration systems.

4

Sztuczne sieci neuronowe jako predykcyjne modele tarcia w układach płynowych

Jankowska A.

Tribologia

|

2003

|

nr 3

81--101

PL

Ze względu na złożony, nieliniowy charakter zjawisk tarciowych, między tłokiem a cylindrem w płynowych układach wykonawczych, za zadowalające uważa się w sterowaniu modele makro - dobrze reprezentujące istotę zjawisk w funkcji ich przyczyn. W pracy przedstawiono neuronowe modele predykcyjne przykładowych zjawisk tarciowych w płynowych układach napędowych. Oszacowano złożoność struktur sztucznych sieci neuronowych (SSN) oraz dokładność predykcji modelowanych zmiennych. Badano modele o postaci wielowarstwowych perceptronów oraz sieci radialnych. Rozważano dwa przykłady: siłę tarcia między nieobciążonym zewnętrzną siłą tłokiem a cylindrem siłownika w pneumatycznym stanowisku badawczym oraz drgania tarciowe w końcowej fazie pozycjonowania tłoka w serwomechanizmie elektrohydraulicznym. W obu przykładach opracowano dość proste struktury SSN realizujące predykcyjne modelowanie z niezłą dokładnością. Niewielka liczba iteracji podczas predykcji z adaptacyjnym dostrajaniem wag sieci wskazuje, że wykorzystanie modelu w trybie on-line staje się możliwe w mikroprocesorowym układzie sterowania. Ze względu na znaczne stałe czasowe płynowych członów wykonawczych, neuronowy model predykcyjny, do pracy bez adaptacji wag w fazie pozycjonowania, czyli zasadniczo w trybie predykcji z długim horyzontem, może się okazać porównywalnie przydatny w układach sterowania jak model z bieżącą korekcją wag.

EN

In the paper the neural models of selected tribological phenomena in fluid drive systems are presented. The complexity of models and modeling accuracy have been estimated. Despite the introduction of standards in friction and wear test stands in many complex technical systems, for example excavators, robots and combines, etc., the fixing conditions of fluid cylinders determine individually the development of frictional vibrations of the piston in a fluid cylinder at low movement velocities. The knowledge of friction models (even if in the form of "black box") presenting the complex nature of tribological phenomena is necessary to achieve desired piston position with a given accuracy. In this paper two examples of tribological phenomena are presented: the friction force between a cylinder and an externally nonloaded piston during the piston braking in the pneumatic research stand, and frictional vibrations in electro-hydraulic servo-drive in final positioning phase. When creating frictional models the Artificial Neural Networks (ANN), characterized by good generalization properties of input data features are applied. Feedforward nets: multilayer (3 or 4 layers) perceptrons (having one or two hidden layers) (MLP) and radial basis functions (RBF), have been tested. The network parameters estimation in learning process realized with the help of gradient optimization methods, was performed. The created predictive models estimate forecasted values of modeled variables on the basis of signal values measured in previous moments. In the first example the friction force as model output was estimated. The inputs were three measured signals: pressure in the sealed chamber, pressure difference at the piston and the velocity of the piston. The quotient of standard deviation of modeling error to pattern standard deviation (quotient) and correlation function (correlation) between model's output and pattern, were estimated. The rather good quality of quotient < or = 0. 2, and correlation > or = 0. 94 with about ten to twenty neurons in hidden layer in all ANN structures created, were obtained. In the second considered example the parameters of frictional vibrations were modeled. The increments of piston position d_x(k) and time lapses between the vibrations d_ T(k), have been estimated. Each parameter served as an output for the three-layer perceptron. The estimation was done on the basis of averaged velocity values in n previous moments of vibrations steps and m previous measured values of piston jumps d_x(k-1), . .„ d_x(k-m) and d_T(k-1), . . ., d_ T(k-m). The number of neurons in net structures and numbers of delays were determined, and n=m=2. An averaged relative prediction error ~1.5% for lapses of time d_T(k) and -4.5% for position jumps d_x(k) in two last points of tested series were obtained in the best MLP 3-layer net of 6:20-7-1structure (6 input samples, 20 conditioning linear neurons, 7 hidden neurons, 1 output neuron). A small number of iteration cycles during adaptive weights tuning in on-line mode of models, indicates the possibility of applying this model in microprocessor control systems. Because of significant time constants of fluid drives the neural predictive models working with multi-step prediction could be applied in control systems - just as models with current adaptive weight tuning. Rather simple ANN structures were compiled for two investigated examples of friction phenomena. Quite good modeling quality was obtained.